register_tsne: Register Seriation Based on 1D t-SNE

In seriation: Infrastructure for Ordering Objects Using Seriation

Register Seriation Based on 1D t-SNE

Description

Use t-distributed stochastic neighbor embedding (t-SNE) for seriate().

Usage

register_tsne()

Details

Registers the method "tsne" for seriate(). This method applies 1D t-SNE to a data matrix or a distance matrix and extracts the order from the 1D embedding. To speed up the process, an initial embedding is created using 1D multi-dimensional scaling (MDS) or principal comonents analysis (PCA) which is improved by t-SNE.

The control parameter "mds" or "pca" controls if MDS (for distances) or PCA (for data matrices) is used to create an initial embedding. See Rtsne::Rtsne() to learn about the other available control parameters.

Perplexity is automatically set as the minimum between 30 and the number of observations. It can be also specified using the control parameter "preplexity".

Note: Package Rtsne needs to be installed.

Value

Nothing.

References

van der Maaten, L.J.P. & Hinton, G.E., 2008. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research, 9, pp.2579-2605.

See Also

Rtsne::Rtsne() in Rtsne.

Other seriation: register_DendSer(), register_GA(), register_optics(), register_smacof(), register_umap(), registry_for_seriaiton_methods, seriate_best(), seriate()

Examples

## Not run: 
register_tsne()

# distances
get_seriation_method("dist", "tsne")

data(SupremeCourt)
d <- as.dist(SupremeCourt)

o <- seriate(d, method = "tsne", verbose = TRUE)
pimage(d, o)

# look at the returned configuration and plot it
attr(o[[1]], "configuration")
plot_config(o)

# the t-SNE results are also available as an attribute (see ? Rtsne::Rtsne)
attr(o[[1]], "model")

## matrix
get_seriation_method("matrix", "tsne")

data("Zoo")
x <- Zoo

x[,"legs"] <- (x[,"legs"] > 0)

# t-SNE does not allow duplicates
x <- x[!duplicated(x), , drop = FALSE]

class <- x$class
label <- rownames(x)
x <- as.matrix(x[,-17])

o <- seriate(x, method = "tsne", eta = 10, verbose = TRUE)
pimage(x, o, prop = FALSE, row_labels = TRUE, col_labels = TRUE)

# look at the row embedding
plot_config(o[[1]], col = class)

## End(Not run)

seriation documentation built on Nov. 27, 2023, 1:07 a.m.